Dev Diary 7

Welcome back! In the last diary, we finished the initial draft of the underlying truth for the first nightmare scenario. Today we will begin working on a market-specific AI. In general, the market-specific AI has two primary goals:

• Provide players with insight into the underlying truth.

• Beat the default strategies common across each market scenario.

Importantly, the market-specific AI is not designed (and usually is not) the clearly optimal trading strategy.

There are two main observations for recapping the underlying truth:

• At similar prices, stocks with a low dividend will have a higher overall return than stocks with a high dividend.

• A ‘bear market’ (or an ~40% decrease) is increasingly likely as the number of stocks at or above $145 increases. A couple of important points on the bear market:

  • A bear market is ~13% likely with 2 stocks, ~40% likely with 3 stocks, and 62% likely with 4 stocks.

  • Stocks below $110 are increasingly less likely to experience the bear market.

Taking these observations together and aiming for a relatively parsimonious strategy, the initial strategy will focus on the number of stocks at or above $145:

• If 1 or fewer stocks is $145 or above: 100% of the portfolio will go into the ABC stock. In this situation, the probability of a bear market is very low, and stocks with a higher dividend would need substantially lower prices to achieve better overall growth (last figure of Dev Diary 5).

• If 2 stocks are $145 or above: 20% of the portfolio will go into the cheapest stock to help hedge the risk of a bear market, 40% of the portfolio will go into the BCUC stock due to the 3rd lowest dividend, and the rest of the portfolio will go into the ABC stock due to the low dividend.

  • Note: There are at least two possible optimizations: (1) only hedge if the cheapest stock is below $110, and/or (2) more carefully allocate between BCUC and ABC based on the observed price differences.

• If 3 stocks are $145 or above: 60% of the portfolio will go into the cheapest stock, 20% of the portfolio will go into the BCUC stock, and the rest will go into the ABC stock.

  • Note: At least 1 more optimization exists, there likely exists an optimal allocation to the cheapest stock (or bonds, if no stocks are below $110) based on the expected overall gain in a bull market (10%) compared to the expected loss in a bear market (40%).

• If 4 or more stocks are $145 or above: 100% of the portfolio will go into the cheapest stock to offset the potential for significant losses.

This strategy is relatively straightforward to implement.

However, the strategy actually fails to beat the default “aggressive AI”, which means it is no better than buying as much stock as possible of a randomly chosen stock on the 1st turn and then thereafter buying the cheapest stock.

To better understand what is going on, we can substitute THC for ABC because it has a lower dividend (but higher variability), and ABC for BCUC. This approach further leans into the general strategy and will tell us if we need a substantial or minor modification.

It actually has worse average performance! One optimization constant across market conditions is buying bonds after executing all of the stock orders.

The bonds optimization has marginally better performance but still fails to beat the aggressive AI. Thus, we will need to implement a more substantial modification: accounting for the price difference between ABC and other stocks. This approach will help get at the core of the underlying truth.

Specifically, several changes were implemented:

• Rather than split invest between BCUC and ABC, stocks are bought based on the relative price between ABC and other stocks. This approach accounts for the possibility that higher dividend stocks may have better overall returns than ABC if their price is sufficiently low.

• To further account for price differences, the stocks of NRI and REAL were also considered. The addition of these stocks further requires ABC to not have an overly expensive price.

Taken together, below is one strategy that follows this approach. Steam post image

And it is better! Given the goals, this strategy can serve as the market-specific AI. It provides insight into the underlying truth while leaving (significant) room for optimization, and it beats the default AIs.

Steam post image Thank you for going through the development of this scenario. In the next diary, we will begin development of the 2nd (out of 3) nightmare scenarios. Talk with you then!